webrtc/sdk/android/api/org/webrtc/YuvConverter.java

/*
 *  Copyright 2015 The WebRTC project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

package org.webrtc;

import android.opengl.GLES11Ext;
import android.opengl.GLES20;
import java.nio.ByteBuffer;
import java.nio.FloatBuffer;
import org.webrtc.VideoFrame.I420Buffer;
import org.webrtc.VideoFrame.TextureBuffer;

/**
 * Class for converting OES textures to a YUV ByteBuffer. It should be constructed on a thread with
 * an active EGL context, and only be used from that thread.
 */
public class YuvConverter {
  // Vertex coordinates in Normalized Device Coordinates, i.e.
  // (-1, -1) is bottom-left and (1, 1) is top-right.
  private static final FloatBuffer DEVICE_RECTANGLE = GlUtil.createFloatBuffer(new float[] {
      -1.0f, -1.0f, // Bottom left.
      1.0f, -1.0f, // Bottom right.
      -1.0f, 1.0f, // Top left.
      1.0f, 1.0f, // Top right.
  });

  // Texture coordinates - (0, 0) is bottom-left and (1, 1) is top-right.
  private static final FloatBuffer TEXTURE_RECTANGLE = GlUtil.createFloatBuffer(new float[] {
      0.0f, 0.0f, // Bottom left.
      1.0f, 0.0f, // Bottom right.
      0.0f, 1.0f, // Top left.
      1.0f, 1.0f // Top right.
  });

  // clang-format off
  private static final String VERTEX_SHADER =
        "varying vec2 interp_tc;\n"
      + "attribute vec4 in_pos;\n"
      + "attribute vec4 in_tc;\n"
      + "\n"
      + "uniform mat4 texMatrix;\n"
      + "\n"
      + "void main() {\n"
      + "    gl_Position = in_pos;\n"
      + "    interp_tc = (texMatrix * in_tc).xy;\n"
      + "}\n";

  private static final String OES_FRAGMENT_SHADER =
        "#extension GL_OES_EGL_image_external : require\n"
      + "precision mediump float;\n"
      + "varying vec2 interp_tc;\n"
      + "\n"
      + "uniform samplerExternalOES tex;\n"
      // Difference in texture coordinate corresponding to one
      // sub-pixel in the x direction.
      + "uniform vec2 xUnit;\n"
      // Color conversion coefficients, including constant term
      + "uniform vec4 coeffs;\n"
      + "\n"
      + "void main() {\n"
      // Since the alpha read from the texture is always 1, this could
      // be written as a mat4 x vec4 multiply. However, that seems to
      // give a worse framerate, possibly because the additional
      // multiplies by 1.0 consume resources. TODO(nisse): Could also
      // try to do it as a vec3 x mat3x4, followed by an add in of a
      // constant vector.
      + "  gl_FragColor.r = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc - 1.5 * xUnit).rgb);\n"
      + "  gl_FragColor.g = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc - 0.5 * xUnit).rgb);\n"
      + "  gl_FragColor.b = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc + 0.5 * xUnit).rgb);\n"
      + "  gl_FragColor.a = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc + 1.5 * xUnit).rgb);\n"
      + "}\n";

  private static final String RGB_FRAGMENT_SHADER =
        "precision mediump float;\n"
      + "varying vec2 interp_tc;\n"
      + "\n"
      + "uniform sampler2D tex;\n"
      // Difference in texture coordinate corresponding to one
      // sub-pixel in the x direction.
      + "uniform vec2 xUnit;\n"
      // Color conversion coefficients, including constant term
      + "uniform vec4 coeffs;\n"
      + "\n"
      + "void main() {\n"
      // Since the alpha read from the texture is always 1, this could
      // be written as a mat4 x vec4 multiply. However, that seems to
      // give a worse framerate, possibly because the additional
      // multiplies by 1.0 consume resources. TODO(nisse): Could also
      // try to do it as a vec3 x mat3x4, followed by an add in of a
      // constant vector.
      + "  gl_FragColor.r = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc - 1.5 * xUnit).rgb);\n"
      + "  gl_FragColor.g = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc - 0.5 * xUnit).rgb);\n"
      + "  gl_FragColor.b = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc + 0.5 * xUnit).rgb);\n"
      + "  gl_FragColor.a = coeffs.a + dot(coeffs.rgb,\n"
      + "      texture2D(tex, interp_tc + 1.5 * xUnit).rgb);\n"
      + "}\n";
  // clang-format on

  private final ThreadUtils.ThreadChecker threadChecker = new ThreadUtils.ThreadChecker();
  private final GlTextureFrameBuffer textureFrameBuffer;
  private TextureBuffer.Type shaderTextureType;
  private GlShader shader;
  private int texMatrixLoc;
  private int xUnitLoc;
  private int coeffsLoc;
  private boolean released = false;

  /**
   * This class should be constructed on a thread that has an active EGL context.
   */
  public YuvConverter() {
    threadChecker.checkIsOnValidThread();
    textureFrameBuffer = new GlTextureFrameBuffer(GLES20.GL_RGBA);
  }

  /** Converts the texture buffer to I420. */
  public I420Buffer convert(TextureBuffer textureBuffer) {
    final int width = textureBuffer.getWidth();
    final int height = textureBuffer.getHeight();

    // SurfaceTextureHelper requires a stride that is divisible by 8.  Round width up.
    // See SurfaceTextureHelper for details on the size and format.
    final int stride = ((width + 7) / 8) * 8;
    final int uvHeight = (height + 1) / 2;
    // Due to the layout used by SurfaceTextureHelper, vPos + stride * uvHeight would overrun the
    // buffer.  Add one row at the bottom to compensate for this.  There will never be data in the
    // extra row, but now other code does not have to deal with v stride * v height exceeding the
    // buffer's capacity.
    final int size = stride * (height + uvHeight + 1);
    ByteBuffer buffer = JniCommon.allocateNativeByteBuffer(size);
    convert(buffer, width, height, stride, textureBuffer.getTextureId(),
        RendererCommon.convertMatrixFromAndroidGraphicsMatrix(textureBuffer.getTransformMatrix()),
        textureBuffer.getType());

    final int yPos = 0;
    final int uPos = yPos + stride * height;
    // Rows of U and V alternate in the buffer, so V data starts after the first row of U.
    final int vPos = uPos + stride / 2;

    buffer.position(yPos);
    buffer.limit(yPos + stride * height);
    ByteBuffer dataY = buffer.slice();

    buffer.position(uPos);
    buffer.limit(uPos + stride * uvHeight);
    ByteBuffer dataU = buffer.slice();

    buffer.position(vPos);
    buffer.limit(vPos + stride * uvHeight);
    ByteBuffer dataV = buffer.slice();

    // SurfaceTextureHelper uses the same stride for Y, U, and V data.
    return JavaI420Buffer.wrap(width, height, dataY, stride, dataU, stride, dataV, stride,
        () -> { JniCommon.freeNativeByteBuffer(buffer); });
  }

  /** Deprecated, use convert(TextureBuffer). */
  @Deprecated
  void convert(ByteBuffer buf, int width, int height, int stride, int srcTextureId,
      float[] transformMatrix) {
    convert(buf, width, height, stride, srcTextureId, transformMatrix, TextureBuffer.Type.OES);
  }

  private void initShader(TextureBuffer.Type textureType) {
    if (shader != null) {
      shader.release();
    }

    final String fragmentShader;
    switch (textureType) {
      case OES:
        fragmentShader = OES_FRAGMENT_SHADER;
        break;
      case RGB:
        fragmentShader = RGB_FRAGMENT_SHADER;
        break;
      default:
        throw new IllegalArgumentException("Unsupported texture type.");
    }

    shaderTextureType = textureType;
    shader = new GlShader(VERTEX_SHADER, fragmentShader);
    shader.useProgram();
    texMatrixLoc = shader.getUniformLocation("texMatrix");
    xUnitLoc = shader.getUniformLocation("xUnit");
    coeffsLoc = shader.getUniformLocation("coeffs");
    GLES20.glUniform1i(shader.getUniformLocation("tex"), 0);
    GlUtil.checkNoGLES2Error("Initialize fragment shader uniform values.");
    // Initialize vertex shader attributes.
    shader.setVertexAttribArray("in_pos", 2, DEVICE_RECTANGLE);
    // If the width is not a multiple of 4 pixels, the texture
    // will be scaled up slightly and clipped at the right border.
    shader.setVertexAttribArray("in_tc", 2, TEXTURE_RECTANGLE);
  }

  private void convert(ByteBuffer buf, int width, int height, int stride, int srcTextureId,
      float[] transformMatrix, TextureBuffer.Type textureType) {
    threadChecker.checkIsOnValidThread();
    if (released) {
      throw new IllegalStateException("YuvConverter.convert called on released object");
    }
    if (textureType != shaderTextureType) {
      initShader(textureType);
    }
    shader.useProgram();

    // We draw into a buffer laid out like
    //
    //    +---------+
    //    |         |
    //    |  Y      |
    //    |         |
    //    |         |
    //    +----+----+
    //    | U  | V  |
    //    |    |    |
    //    +----+----+
    //
    // In memory, we use the same stride for all of Y, U and V. The
    // U data starts at offset |height| * |stride| from the Y data,
    // and the V data starts at at offset |stride/2| from the U
    // data, with rows of U and V data alternating.
    //
    // Now, it would have made sense to allocate a pixel buffer with
    // a single byte per pixel (EGL10.EGL_COLOR_BUFFER_TYPE,
    // EGL10.EGL_LUMINANCE_BUFFER,), but that seems to be
    // unsupported by devices. So do the following hack: Allocate an
    // RGBA buffer, of width |stride|/4. To render each of these
    // large pixels, sample the texture at 4 different x coordinates
    // and store the results in the four components.
    //
    // Since the V data needs to start on a boundary of such a
    // larger pixel, it is not sufficient that |stride| is even, it
    // has to be a multiple of 8 pixels.

    if (stride % 8 != 0) {
      throw new IllegalArgumentException("Invalid stride, must be a multiple of 8");
    }
    if (stride < width) {
      throw new IllegalArgumentException("Invalid stride, must >= width");
    }

    int y_width = (width + 3) / 4;
    int uv_width = (width + 7) / 8;
    int uv_height = (height + 1) / 2;
    int total_height = height + uv_height;
    int size = stride * total_height;

    if (buf.capacity() < size) {
      throw new IllegalArgumentException("YuvConverter.convert called with too small buffer");
    }
    // Produce a frame buffer starting at top-left corner, not
    // bottom-left.
    transformMatrix =
        RendererCommon.multiplyMatrices(transformMatrix, RendererCommon.verticalFlipMatrix());

    final int frameBufferWidth = stride / 4;
    final int frameBufferHeight = total_height;
    textureFrameBuffer.setSize(frameBufferWidth, frameBufferHeight);

    // Bind our framebuffer.
    GLES20.glBindFramebuffer(GLES20.GL_FRAMEBUFFER, textureFrameBuffer.getFrameBufferId());
    GlUtil.checkNoGLES2Error("glBindFramebuffer");

    GLES20.glActiveTexture(GLES20.GL_TEXTURE0);
    GLES20.glBindTexture(textureType.getGlTarget(), srcTextureId);
    GLES20.glUniformMatrix4fv(texMatrixLoc, 1, false, transformMatrix, 0);

    // Draw Y
    GLES20.glViewport(0, 0, y_width, height);
    // Matrix * (1;0;0;0) / width. Note that opengl uses column major order.
    GLES20.glUniform2f(xUnitLoc, transformMatrix[0] / width, transformMatrix[1] / width);
    // Y'UV444 to RGB888, see
    // https://en.wikipedia.org/wiki/YUV#Y.27UV444_to_RGB888_conversion.
    // We use the ITU-R coefficients for U and V */
    GLES20.glUniform4f(coeffsLoc, 0.299f, 0.587f, 0.114f, 0.0f);
    GLES20.glDrawArrays(GLES20.GL_TRIANGLE_STRIP, 0, 4);

    // Draw U
    GLES20.glViewport(0, height, uv_width, uv_height);
    // Matrix * (1;0;0;0) / (width / 2). Note that opengl uses column major order.
    GLES20.glUniform2f(
        xUnitLoc, 2.0f * transformMatrix[0] / width, 2.0f * transformMatrix[1] / width);
    GLES20.glUniform4f(coeffsLoc, -0.169f, -0.331f, 0.499f, 0.5f);
    GLES20.glDrawArrays(GLES20.GL_TRIANGLE_STRIP, 0, 4);

    // Draw V
    GLES20.glViewport(stride / 8, height, uv_width, uv_height);
    GLES20.glUniform4f(coeffsLoc, 0.499f, -0.418f, -0.0813f, 0.5f);
    GLES20.glDrawArrays(GLES20.GL_TRIANGLE_STRIP, 0, 4);

    GLES20.glReadPixels(
        0, 0, frameBufferWidth, frameBufferHeight, GLES20.GL_RGBA, GLES20.GL_UNSIGNED_BYTE, buf);

    GlUtil.checkNoGLES2Error("YuvConverter.convert");

    // Restore normal framebuffer.
    GLES20.glBindFramebuffer(GLES20.GL_FRAMEBUFFER, 0);
    GLES20.glBindTexture(GLES20.GL_TEXTURE_2D, 0);

    // Unbind texture. Reportedly needed on some devices to get
    // the texture updated from the camera.
    GLES20.glBindTexture(textureType.getGlTarget(), 0);
  }

  public void release() {
    threadChecker.checkIsOnValidThread();
    released = true;
    if (shader != null) {
      shader.release();
    }
    textureFrameBuffer.release();
  }
}